今天跟大家分享一篇语音相关的论文合集：cs.SD语音2篇，eess.AS音频处理4篇。
【1】 TB or not TB? Acoustic cough analysis for tuberculosis classification
标题：结核病还是非结核病？声学咳嗽分析在结核病分类中的应用
链接：https://arxiv.org/abs/2209.00934
作者：Geoffrey Frost,Grant Theron,Thomas Niesler机构：Department of Electrical and Electronic Engineering, Stellenbosch University, South Africa, SAMRC Centre for Tuberculosis Research, DSINRF Centre of Excellence for Biomedical, Tuberculosis Research, Stellenbosch University, South Africa备注：Accepted for publication at Interspeech 2022摘要：在这项工作中，我们探索结核病（TB）咳嗽分类的递归神经网络结构。与之前在该领域实现深度架构的不成功尝试相比，我们展示了基本的双向长短期存储器网络（BiLSTM）可以实现改进的性能。此外，我们还表明，通过执行贪婪特征选择，结合新提出的基于注意力的架构（学习患者不变特征），与基线和其他考虑的架构相比，可以实现实质上更好的泛化。此外，该注意机制允许检查被认为对于要执行的分类是重要的音频信号的时间区域。最后，我们开发了一种神经风格转移技术来推断理想化的输入，这些输入可以随后进行分析。我们发现理想化的肺结核咳嗽和非肺结核咳嗽的功率谱之间存在明显差异，这提供了关于音频信号中特征起源的线索。摘要：In this work, we explore recurrent neural network architectures for tuberculosis (TB) cough classification. In contrast to previous unsuccessful attempts to implement deep architectures in this domain, we show that a basic bidirectional long short-term memory network (BiLSTM) can achieve improved performance. In addition, we show that by performing greedy feature selection in conjunction with a newly-proposed attention-based architecture that learns patient invariant features, substantially better generalisation can be achieved compared to a baseline and other considered architectures. Furthermore, this attention mechanism allows an inspection of the temporal regions of the audio signal considered to be important for classification to be performed. Finally, we develop a neural style transfer technique to infer idealised inputs which can subsequently be analysed. We find distinct differences between the idealised power spectra of TB and non-TB coughs, which provide clues about the origin of the features in the audio signal.
【2】 A Wavelet Transform Based Scheme to Extract Speech Pitch and Formant Frequencies
标题：一种基于小波变换的语音基音和共振峰频率提取方法
链接：https://arxiv.org/abs/2209.00733
作者：Mahdieh Ghazvini,Seyedamiryousef Hosseini Goki,Sajad Hamzenejadi机构：Department of Computer Engineering, Shahid Bahonar University of Kerman, Kerman, Iran, Department of Computer Science, University of Victoria, Victoria, BC, Canada, Department of Industrial and Information Engineering, Politecnico di Milano, Milan, Italy摘要：基音频率和共振峰频率是语音处理应用中的重要特征。声带输出元音的周期被称为音高或基频，共振峰频率基本上是声道的共振频率。这些特征在不同的人甚至不同的词之间是不同的，但它们都在一定的频率范围内。在实践中，对于大多数语音处理来说，仅仅前三个共振峰就足够了。特征提取和分类是每个语音识别系统的主要组成部分。本文提出了两种基于小波变换的特征提取方法，并利用滤波器组的思想对这两种特征进行了提取。通过对几种语音信号的特征提取结果进行比较，发现小波变换比倒谱方法具有更好的准确性，且对噪声不敏感。此外，还对几种基于模糊分类的语音处理技术进行了综述。摘要：Pitch and Formant frequencies are important features in speech processing applications. The period of the vocal cord's output for vowels is known as the pitch or the fundamental frequency, and formant frequencies are essentially resonance frequencies of the vocal tract. These features vary among different persons and even words, but they are within a certain frequency range. In practice, just the first three formants are enough for the most of speech processing. Feature extraction and classification are the main components of each speech recognition system. In this article, two wavelet based approaches are proposed to extract the mentioned features with help of the filter bank idea. By comparing the results of the presented feature extraction methods on several speech signals, it was found out that the wavelet transform has a good accuracy compared to the cepstrum method and it has no sensitivity to noise. In addition, several fuzzy based classification techniques for speech processing are reviewed.
【1】 Inverse-free Online Independent Vector Analysis with Flexible Iterative Source Steering
标题：灵活迭代信源控制的无逆在线独立向量分析
链接：https://arxiv.org/abs/2209.00937
作者：Taishi Nakashima,Nobutaka Ono机构：Tokyo Metropolitan University, Tokyo, Japan.备注：5 pages, 2 figures. Submitted to APSIPA 2022摘要：提出了一种新的在线独立矢量分析（IVA）算法用于实时盲源分离（BSS）。在许多盲源分离算法中，迭代投影（IP）被用于更新分层矩阵，分层矩阵是盲源分离中要估计的参数。然而，它需要矩阵求逆，这可能是昂贵的，特别是在在线处理中。为了改善这种情况，我们将迭代源转向（ISS）引入在线IVA。ISS不需要任何矩阵求逆，因此其计算复杂度低于IP。此外，当只有部分信源运动时，ISS算法可以灵活有效地更新分层矩阵，使得只有运动信源的导向矢量得到更新。动态条件下的数值实验验证了所提方法的有效性。摘要：In this paper, we propose a new online independent vector analysis (IVA) algorithm for real-time blind source separation (BSS). In many BSS algorithms, the iterative projection (IP) has been used for updating the demixing matrix, a parameter to be estimated in BSS. However, it requires matrix inversion, which can be costly, particularly in online processing. To improve this situation, we introduce iterative source steering (ISS) to online IVA. ISS does not require any matrix inversions, and thus its computational complexity is less than that of IP. Furthermore, when only part of the sources are moving, ISS enables us to update the demixing matrix flexibly and effectively so that the steering vectors of only the moving sources are updated. Numerical experiments under a dynamic condition confirm the efficacy of the proposed method.
【2】 TB or not TB? Acoustic cough analysis for tuberculosis classification
标题：结核病还是非结核病？声学咳嗽分析在结核病分类中的应用
链接：https://arxiv.org/abs/2209.00934
* 与cs.SD语音【1】为同一篇
作者：Geoffrey Frost,Grant Theron,Thomas Niesler机构：Department of Electrical and Electronic Engineering, Stellenbosch University, South Africa, SAMRC Centre for Tuberculosis Research, DSINRF Centre of Excellence for Biomedical, Tuberculosis Research, Stellenbosch University, South Africa备注：Accepted for publication at Interspeech 2022摘要：在这项工作中，我们探索结核病（TB）咳嗽分类的递归神经网络结构。与之前在该领域实现深度架构的不成功尝试相比，我们展示了基本的双向长短期存储器网络（BiLSTM）可以实现改进的性能。此外，我们还表明，通过执行贪婪特征选择，结合新提出的基于注意力的架构（学习患者不变特征），与基线和其他考虑的架构相比，可以实现实质上更好的泛化。此外，该注意机制允许检查被认为对于要执行的分类是重要的音频信号的时间区域。最后，我们开发了一种神经风格转移技术来推断理想化的输入，这些输入可以随后进行分析。我们发现理想化的肺结核咳嗽和非肺结核咳嗽的功率谱之间存在明显差异，这提供了关于音频信号中特征起源的线索。摘要：In this work, we explore recurrent neural network architectures for tuberculosis (TB) cough classification. In contrast to previous unsuccessful attempts to implement deep architectures in this domain, we show that a basic bidirectional long short-term memory network (BiLSTM) can achieve improved performance. In addition, we show that by performing greedy feature selection in conjunction with a newly-proposed attention-based architecture that learns patient invariant features, substantially better generalisation can be achieved compared to a baseline and other considered architectures. Furthermore, this attention mechanism allows an inspection of the temporal regions of the audio signal considered to be important for classification to be performed. Finally, we develop a neural style transfer technique to infer idealised inputs which can subsequently be analysed. We find distinct differences between the idealised power spectra of TB and non-TB coughs, which provide clues about the origin of the features in the audio signal.
【3】 Multi-scale temporal-frequency attention for music source separation
标题：用于音乐源分离的多尺度时频注意
链接：https://arxiv.org/abs/2209.00805
作者：Lianwu Chen,Xiguang Zheng,Chen Zhang,Liang Guo,Bing Yu机构：Kuaishou Technology Co. Beijing, China摘要：近年来，基于深度神经网络（DNNs）的音乐源分离（MSS）方法已经取得了最新的性能。虽然先前的方法已经使用各种方法解决了大感受野建模，但是对于MSS任务，具有重复模式的音乐谱图的时间和频率相关性还没有被明确地探索。本文提出了一个时频注意模型，用于对语谱图在时间和频率两个维度上的相关性进行建模。此外，提出了一种多尺度关注度算法，有效地捕捉音乐信号的相关性。在MUSDB18数据集上的实验结果表明，该方法在分离声干方面的性能优于现有系统，信噪比为9.51dB，是MSS的主要实际应用。摘要：In recent years, deep neural networks (DNNs) based approaches have achieved the start-of-the-art performance for music source separation (MSS). Although previous methods have addressed the large receptive field modeling using various methods, the temporal and frequency correlations of the music spectrogram with repeated patterns have not been explicitly explored for the MSS task. In this paper, a temporal-frequency attention module is proposed to model the spectrogram correlations along both temporal and frequency dimensions. Moreover, a multi-scale attention is proposed to effectively capture the correlations for music signal. The experimental results on MUSDB18 dataset show that the proposed method outperforms the existing state-of-the-art systems with 9.51 dB signal-to-distortion ratio (SDR) on separating the vocal stems, which is the primary practical application of MSS.
【4】 A Wavelet Transform Based Scheme to Extract Speech Pitch and Formant Frequencies
标题：一种基于小波变换的语音基音和共振峰频率提取方法
链接：https://arxiv.org/abs/2209.00733
* 与cs.SD语音【2】为同一篇
作者：Mahdieh Ghazvini,Seyedamiryousef Hosseini Goki,Sajad Hamzenejadi机构：Department of Computer Engineering, Shahid Bahonar University of Kerman, Kerman, Iran, Department of Computer Science, University of Victoria, Victoria, BC, Canada, Department of Industrial and Information Engineering, Politecnico di Milano, Milan, Italy摘要：基音频率和共振峰频率是语音处理应用中的重要特征。声带输出元音的周期被称为音高或基频，共振峰频率基本上是声道的共振频率。这些特征在不同的人甚至不同的词之间是不同的，但它们都在一定的频率范围内。在实践中，对于大多数语音处理来说，仅仅前三个共振峰就足够了。特征提取和分类是每个语音识别系统的主要组成部分。本文提出了两种基于小波变换的特征提取方法，并利用滤波器组的思想对这两种特征进行了提取。通过对几种语音信号的特征提取结果进行比较，发现小波变换比倒谱方法具有更好的准确性，且对噪声不敏感。此外，还对几种基于模糊分类的语音处理技术进行了综述。摘要：Pitch and Formant frequencies are important features in speech processing applications. The period of the vocal cord's output for vowels is known as the pitch or the fundamental frequency, and formant frequencies are essentially resonance frequencies of the vocal tract. These features vary among different persons and even words, but they are within a certain frequency range. In practice, just the first three formants are enough for the most of speech processing. Feature extraction and classification are the main components of each speech recognition system. In this article, two wavelet based approaches are proposed to extract the mentioned features with help of the filter bank idea. By comparing the results of the presented feature extraction methods on several speech signals, it was found out that the wavelet transform has a good accuracy compared to the cepstrum method and it has no sensitivity to noise. In addition, several fuzzy based classification techniques for speech processing are reviewed.
机器翻译，仅供参考
每日语音速递